【作者】 李海霞；

【导师】 朱德海； 李保国；

【作者基本信息】 中国农业大学 ， 土地利用与信息技术， 2014， 博士

【摘要】 随着遥感技术的发展,更高空间分辨率的航天遥感数据产品不断出现,以及航空遥感技术的日臻成熟,为城乡规划遥感监测提供了有效而适用的数据源。本文以WorldView-Ⅱ卫星遥感数据结合高分辨率的航空遥感数据,针对我国迅速城市化过程中城乡规划与建设的监测问题,开展适合高空间分辨率遥感影像的面向对象特征构建、特征选择和城市地物分类等技术研究,并进一步应用于城乡规划的控制性详细规划监测。主要研究内容与结果如下：(1)在影像多尺度分割生成对象的基础上,研究了对象特征构建算法,重点研究了城市地物较突出的纹理特征和形状特征构建算法。首先研究了基于半变异函数的纹理计算尺度问题,并进一步研究了基于灰度共生矩阵的对象纹理特征计算算法。针对城市建筑地物形状规整的特点,研究了多种形状特征计算算法。(2)分割后的高分遥感影像对象有光谱、纹理和形状等多种特征的特点,众多特征参与分类,分类效率会降低,但分类精度也未必能提高。针对面向对象分类有益特征选择问题,研究了对象特征选择的技术流程,首先基于改进的ReliefF去除无关特征,再利用特征间的互信息初步去除冗余特征,形成特征粗集,最后利用遗传算法,以类内、类间距离为适应度函数,对影像对象分类的最优特征子集进行搜索和评价,最终得到有益于城乡地物对象分类的特征子集。(3)针对面向对象分类的小样本问题,结合SVM适合小样本的特点,本文针对基于SVM的面向对象分类中的部分环节进行了研究。针对径向基核函数SVM的C和γ参数对分类精度影响大的问题,在前人研究的基础上,结合城市监测的需要,研究了三步搜索法的参数优化算法。SVM是针对二分类问题,本文设计了多类分类的流程,即在投票法SVM分类的基础上,进一步利用最近邻法,对票数相同或相差一票的情况进一步分类。最后,以北京市某2个区域为研究区,以高分辨率的WorldView-Ⅱ航天卫星遥感影像为数据源,辅以研究区航空遥感影像,构建遥感城乡土地利用分类体系。在影像分割的基础上,利用本文研究的特征构建算法、特征优选技术流程,以及基于改进SVM分类算法,完成研究区的城市地类分类。最后以研究区的城市规划图为依据,对该区域的城市规划执行情况进行监测和评价。更多还原

【Abstract】 With the development of the remote sensing (RS) technology, the high spatial resolution satellite remote sensing data are presented and the aerial remote sensing technology has been full developed, which provided good data source for the RS monitoring of the urban and rural plan. So the WorldView-Ⅱ data and the high spatial resolution aerial RS image are used in this paper. To the need of the monitoring and management of urban and rural, the object features computing algorithm and feature selection algorithm and classification based object-oriented (OO) methods for urban and rural land use are studied and used in the urban and rural plan monitoring. The main contents and the results are as following:Based on the high spatial remote sensing image segmentation algorithm, the feature computeing algorithm are studied. The obvious texture and shape features of the city object are focused on The scale of texture computing are studied based on semivariogram and the object texture computing algorithm are studied basen grey-level co-occurrence matrix. To the characteristic of the city building objects, several shape features computeing are studied in this paper.There are many features of the object such as spectral features, texture features, shape features and so on. The classification effective will reduce and the classification precision may not be improved with the total features of the objects. To the feature selection of the object-oriented classification, the technology flow of feature selection is formed firstly. The enhanced ReliefF algorithm is adopted to filter the irrelevant features. Then the mutual information among features is computed to eliminate the redundant features. At last, the genetic algorithm is used and the distance between the same classes and the different classes acted as the fitness function.the features set are searched and evaluated, which is beneficial to the city land use classifcication.To the problem of the small sample size in the object-oriented classification, and the support vector machine (SVM) is fit to the small sample size problem, parts of application SVM in the OO classification are studied in this paper. The classification precision are severely impacted by the C and γ parameters in SVM based radial basis function. Based on the research of the former and combined with the need of the urban and rural plan monitoring and management, the three step search method are used to parameters optimization. The SVM is to the two classes, so the multiple classes flow is designed in this paper. After classification based on the vote, further the Nearest Neighhood (NN) classifier are used to the case of vote equiation and one difference vote.Two area in Beijing are the study area and the WorldView-Ⅱ data and the aerial RS image with0.2m spatial resolution are used as the main data sources. The classification system of the urban and rural land use with RS data are designed. Based on the image segmentation and the features computeing and features selection algorithm presented in this paper, then the urban and rural land use is classified with the improved SVM classification. At last comparing with the urban and rural plan map of the area, the urban and rural plan are monitored and evaluated.更多还原